skiwi

Rayz2016 said:

rotateleftbyte said:

red oak said:

Fun seeing washed up “ consultants” out there trying to push back the sea of Apple Silicon performance that is going to wash over x86

Some are saying that the M1 chip is not an Apple design at all and that they 'bought' it in. Oh well, it takes all sorts does it not?

Well, if it makes them feel better about things then they can go right ahead believing that.

Not really going to make any difference though is it?

I guess what they could do is start GoFundMe campaign to raise money to pay someone to claim that they've been doing all Apple's chip design from their fold-down table in the dining room.

The first true Apple design was the A4.  This is the 8th generation of 64 bit design. They have led the mobile SOC space since at least the a5, and the A7 (the first 64bit design) was years ahead of anyone else.

This is all Apple, one of the giveaways is how secretive they are, and also how much they stress "SOC" performance - they are not building discrete CPUs as their competitors do (noting that AMD and Intel have no choice due to the constraints with the ISA), they are building SOC with as much specialisation and as much integration as they can.  

Finally, of course Apple's 5nm M1 is going to outperform Intel's 14nm Core i5. But when Intel's Core i5 is also 14nm in about 3 years (if they hire TSMC to make the chips) or 5 years (if they make the chips themselves)? Then we will see whose performance will wash over whose. Apple will have some advantages, namely the inherent efficiency of RISC vs ISA as well as Apple's strategy of maximizing performance from each core as opposed to Intel's - and everyone else's - strategy of maximizing cores. But Intel also has a performance attribute of their own: the densest core design in the industry. Meaning that a 10nm Intel design is the equivalent of a 7nm Apple one. So when Intel does get to a 5nm design, it will be the equivalent of a 3nm Apple one. So, we shall see ... 

Could have, should have, would have...

Fact is that Intel has spent the last 5 years stagnating and both AMD and now Apple have called them out on it.  You might like your Intel shares, but it doesn't mean that the rest of us have to...

And, btw, the microarchitecture of this 8th generation Apple part blows Intel out of the water.  8-wide decode, 6 times the L1 cache (at lower latency), twice the re-order buffer, etc etc.  They have been able to get away with a much wider design than Intel and still drive it to high clock speeds upending the conventional wisdom that you can only get high clock rates with a narrow design.  Granted the M1 will likely not top 3.5GHz, (obvioulsy the MBA will be lower than the MBP), but doing that with such a wide design is the secret of their significant horsepower.  Also it will undoubtedly be true that, granted the limitations of the memory @16GB, the wide and fast access to memory (and to GPU) will also be a significant factor in performance.

For Intel, the train has left the station and they are trying to run after it....

k2kw said:

beeble42 said:

I'd like to see actual benchmark results, especially for graphics performance, including against a mac mini with an egpu with a reasonable card in it, like a vega64 or something. Saying the integrated graphics are 6 times faster than the previous intel one is fine, but that isn't a particularly high bar when you're removing any option of more powerful gpu technology which the previous one had. The new integrated gpu is competing (from a performance perspective) against the fastest gpu you could get in an egpu box that was supported by the previous model. I doubt the new model is actually faster than that, but it may well be fast enough to beat a moderate egpu setup, and without the expense, meaning a win for Apple. Or maybe it isn't and people will wait longer to upgrade until performance catches up to what they're leaving behind. Or switch platforms.

Without the eGPU support this feels unfinished.   More like 0.8 version.   And why can’t they have 32 or 64 GB RAM .   Step backwards.   Are they trying for it to be no so successful.

You're overthinking it.  And you are stuck with your ISA glasses on.  Unified memory is just a buzzword, but what it actually means is the the memory interface is not nearly as constrained as with the ISA, both in width and depth.  Ditto the GPU.

This is a deeply impressive piece of silicon. Granted it's an 8th-generation design (since the 64 bit Cyclone), but the Firestorm microarchitecture is extremely impressive compared to those of AMD and Intel.  A few examples:
1) 8-wide decode block (2x AMD, Intel a 1-4); 
2) 192KB L1 cache (6x Intel , 3x AMD), with a 3-cycle latency (Intel 5-cycle, AMD, 4-cycle);
3) Re-order Buffer; 600+ instructions (!!!) = Intel Sunny Cove 320, AMD Zen 3 - 256;
4) 4x FADDs and 4x FMULs per cycle with 3 and 4 cycles latency. 4x Intel, 2x Zen 3.

There is a bunch of other stuff that the tech sites are unpicking, but as I said, Apple would not BS the numbers, as they have WAY too much to lose.  I expect that you will find the opposite in actual fact, that the chip performers better than expected.  Yes, really.

Note to the author of the article, you need to make clear that the M1 chip is NOT the same in the MBA as the Mini.  The use of the fan in the MBP and Mini speaks to a different implementation philosophy so you would expect that the M1 is engineered in the MBP and Mini to have higher wattage, which means that the performance will be better on the cooled platforms.

"In theory, Apple could increase the clock speed of the M1 for higher computation performance with the same number of cores. In practice, power usage (and waste heat) grows in a nonlinear relationship to clock speed, so it’d be more like an Intel processor for power usage."

Correct, but also important to remember that this is actually Apple's 8th generation of 64bit microarchitecture (since the Cyclone A7 core).  The microarchitecture of the M1 is very impressive, from what little information can be gleaned by specialist sites (note Apple don't disclose anything about it).  As an example, leveraging the fixed length instruction set of ARM, they run an 8-wide decode which is twice what anyone else is doing on Arm, AMD and Intel.  

In "old money" that's like running a 4-barrel Holley into your small-block over and above the old 2-barrel Rochester.

And, the other interesting thing about this chip over the A14 (which also uses 5nm) is the almost 50% increase in transistors. That is like an upgrade from a v6 to a v8 in old money.
;-)

entropys said:

2morrow said:

Could some explain why they were not able to update the higher end MacBook Pros with the M1?

Because:

• the M1 is currently Apple’slow end chip. That it is being compared with, and can compete with,  higher end alternative chips is just marketing gravy.

That clearly sells the chip short.  The M1 is quite simply the highest performing laptop CPU in the world. By some way (looks like approx 50% over a 10-gen i9).

That said, the following limitations means that this won't be powering the higher-end laptops without another SOC package
1) Limitations in Thunderbolt channels
2) SOC memory capacity limited to 16GB
3) Monitor support limited to 1 addition.